← Agents

Internal knowledge base

An agent keeps Wilow's product knowledge base true to what actually shipped.

How it works

GitHubRead what shipped

The knowledge agent starts from the code that shipped that day. Code proves something shipped; a closed ticket doesn't.

ClickUpMatch to the task

The knowledge agent pairs each change with its ClickUp task to learn what it was meant to do. Unmatched changes go to a review queue.

LarkRead meeting notes

The knowledge agent reads summaries of the team's recorded calls for the reasoning behind each change.

SlackScan team chat

The knowledge agent reads the team's Slack channel for decisions made since its last run.

ClaudeUpdate the docs

The knowledge agent rewrites one doc section per feature and marks how sure it is: unconfirmed, shipped in code, or confirmed in the app.

VercelPublish to the wiki

The knowledge agent saves to the team wiki, and never overwrites a teammate's edit.

WebflowCorrect the website

The knowledge agent checks trywilow.com against what shipped, fixes outdated claims itself and drafts the rest for me.

What went into it

  • Shipped code is the source of truth. A closed task doesn't count as proof.
  • Decisions also go to the team wiki, gathered from meetings, Slack and task comments.
  • An early version reported success on work it never finished, so it was rebuilt to do the work itself.
  • The website check exists because an audit found 9 live pages denying shipped features, on pages bringing in about 65% of search clicks.

Built with

GitHubClickUpLarkSlackClaudeVercelWebflow